20
33
that has merged into it. As the first step of a merge, pull, or
21
34
branch operation we copy history from the source into the destination
37
The copying is done in a slightly complicated order. We don't want to
38
add a revision to the store until everything it refers to is also
39
stored, so that if a revision is present we can totally recreate it.
40
However, we can't know what files are included in a revision until we
41
read its inventory. Therefore, we first pull the XML and hold it in
42
memory until we've updated all of the files referenced.
25
from __future__ import absolute_import
29
from bzrlib.lazy_import import lazy_import
30
lazy_import(globals(), """
41
from bzrlib.i18n import gettext
42
from bzrlib.revision import NULL_REVISION
43
from bzrlib.trace import mutter
46
class RepoFetcher(object):
47
"""Pull revisions and texts from one repository to another.
49
This should not be used directly, it's essential a object to encapsulate
50
the logic in InterRepository.fetch().
45
# TODO: Avoid repeatedly opening weaves so many times.
47
# XXX: This doesn't handle ghost (not present in branch) revisions at
48
# all yet. I'm not sure they really should be supported.
50
# NOTE: This doesn't copy revisions which may be present but not
51
# merged into the last revision. I'm not sure we want to do that.
53
# - get a list of revisions that need to be pulled in
54
# - for each one, pull in that revision file
55
# and get the inventory, and store the inventory with right
57
# - and get the ancestry, and store that with right parents too
58
# - and keep a note of all file ids and version seen
59
# - then go through all files; for each one get the weave,
60
# and add in all file versions
64
def greedy_fetch(to_branch, from_branch, revision=None, pb=None):
65
f = Fetcher(to_branch, from_branch, revision, pb)
66
return f.count_copied, f.failed_revisions
70
class Fetcher(object):
71
"""Pull revisions and texts from one branch to another.
73
This doesn't update the destination's history; that can be done
74
separately if desired.
77
If set, pull only up to this revision_id.
81
last_revision -- if last_revision
82
is given it will be that, otherwise the last revision of
85
count_copied -- number of revisions copied
87
count_weaves -- number of file weaves copied
53
def __init__(self, to_repository, from_repository, last_revision=None,
54
find_ghosts=True, fetch_spec=None):
55
"""Create a repo fetcher.
57
:param last_revision: If set, try to limit to the data this revision
59
:param fetch_spec: A SearchResult specifying which revisions to fetch.
60
If set, this overrides last_revision.
61
:param find_ghosts: If True search the entire history for ghosts.
63
# repository.fetch has the responsibility for short-circuiting
64
# attempts to copy between a repository and itself.
65
self.to_repository = to_repository
66
self.from_repository = from_repository
67
self.sink = to_repository._get_sink()
68
# must not mutate self._last_revision as its potentially a shared instance
69
self._last_revision = last_revision
70
self._fetch_spec = fetch_spec
71
self.find_ghosts = find_ghosts
72
self.from_repository.lock_read()
73
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
74
self.from_repository, self.from_repository._format,
75
self.to_repository, self.to_repository._format)
79
self.from_repository.unlock()
82
"""Primary worker function.
84
This initialises all the needed variables, and then fetches the
85
requested revisions, finally clearing the progress bar.
87
# Roughly this is what we're aiming for fetch to become:
89
# missing = self.sink.insert_stream(self.source.get_stream(search))
91
# missing = self.sink.insert_stream(self.source.get_items(missing))
89
def __init__(self, to_branch, from_branch, last_revision=None, pb=None):
90
if to_branch == from_branch:
91
raise Exception("can't fetch from a branch to itself")
92
self.to_branch = to_branch
93
self.to_weaves = to_branch.weave_store
94
self.to_control = to_branch.control_weaves
95
self.from_branch = from_branch
96
self.from_weaves = from_branch.weave_store
97
self.from_control = from_branch.control_weaves
98
self.failed_revisions = []
93
100
self.count_total = 0
94
self.file_ids_names = {}
95
pb = ui.ui_factory.nested_progress_bar()
96
pb.show_pct = pb.show_count = False
98
pb.update(gettext("Finding revisions"), 0, 2)
99
search_result = self._revids_to_fetch()
100
mutter('fetching: %s', search_result)
101
if search_result.is_empty():
103
pb.update(gettext("Fetching revisions"), 1, 2)
104
self._fetch_everything_for_search(search_result)
108
def _fetch_everything_for_search(self, search):
109
"""Fetch all data for the given set of revisions."""
110
# The first phase is "file". We pass the progress bar for it directly
111
# into item_keys_introduced_by, which has more information about how
112
# that phase is progressing than we do. Progress updates for the other
113
# phases are taken care of in this function.
114
# XXX: there should be a clear owner of the progress reporting. Perhaps
115
# item_keys_introduced_by should have a richer API than it does at the
116
# moment, so that it can feed the progress information back to this
118
if (self.from_repository._format.rich_root_data and
119
not self.to_repository._format.rich_root_data):
120
raise errors.IncompatibleRepositories(
121
self.from_repository, self.to_repository,
122
"different rich-root support")
123
pb = ui.ui_factory.nested_progress_bar()
125
pb.update("Get stream source")
126
source = self.from_repository._get_source(
127
self.to_repository._format)
128
stream = source.get_stream(search)
129
from_format = self.from_repository._format
130
pb.update("Inserting stream")
131
resume_tokens, missing_keys = self.sink.insert_stream(
132
stream, from_format, [])
134
pb.update("Missing keys")
135
stream = source.get_stream_for_missing_keys(missing_keys)
136
pb.update("Inserting missing keys")
137
resume_tokens, missing_keys = self.sink.insert_stream(
138
stream, from_format, resume_tokens)
140
raise AssertionError(
141
"second push failed to complete a fetch %r." % (
144
raise AssertionError(
145
"second push failed to commit the fetch %r." % (
147
pb.update("Finishing stream")
152
def _revids_to_fetch(self):
153
"""Determines the exact revisions needed from self.from_repository to
154
install self._last_revision in self.to_repository.
156
:returns: A SearchResult of some sort. (Possibly a
157
PendingAncestryResult, EmptySearchResult, etc.)
159
if self._fetch_spec is not None:
160
# The fetch spec is already a concrete search result.
161
return self._fetch_spec
162
elif self._last_revision == NULL_REVISION:
163
# fetch_spec is None + last_revision is null => empty fetch.
164
# explicit limit of no revisions needed
165
return vf_search.EmptySearchResult()
166
elif self._last_revision is not None:
167
return vf_search.NotInOtherForRevs(self.to_repository,
168
self.from_repository, [self._last_revision],
169
find_ghosts=self.find_ghosts).execute()
170
else: # self._last_revision is None:
171
return vf_search.EverythingNotInOther(self.to_repository,
172
self.from_repository,
173
find_ghosts=self.find_ghosts).execute()
176
class Inter1and2Helper(object):
177
"""Helper for operations that convert data from model 1 and 2
179
This is for use by fetchers and converters.
182
# This is a class variable so that the test suite can override it.
183
known_graph_threshold = 100
185
def __init__(self, source):
188
:param source: The repository data comes from
192
def iter_rev_trees(self, revs):
193
"""Iterate through RevisionTrees efficiently.
195
Additionally, the inventory's revision_id is set if unset.
197
Trees are retrieved in batches of 100, and then yielded in the order
200
:param revs: A list of revision ids
202
# In case that revs is not a list.
205
for tree in self.source.revision_trees(revs[:100]):
206
if tree.root_inventory.revision_id is None:
207
tree.root_inventory.revision_id = tree.get_revision_id()
211
def _find_root_ids(self, revs, parent_map, graph):
213
for tree in self.iter_rev_trees(revs):
214
root_id = tree.get_root_id()
215
revision_id = tree.get_file_revision(root_id, u"")
216
revision_root[revision_id] = root_id
217
# Find out which parents we don't already know root ids for
219
for revision_parents in parent_map.itervalues():
220
parents.update(revision_parents)
221
parents.difference_update(revision_root.keys() + [NULL_REVISION])
222
# Limit to revisions present in the versionedfile
223
parents = graph.get_parent_map(parents).keys()
224
for tree in self.iter_rev_trees(parents):
225
root_id = tree.get_root_id()
226
revision_root[tree.get_revision_id()] = root_id
229
def generate_root_texts(self, revs):
230
"""Generate VersionedFiles for all root ids.
232
:param revs: the revisions to include
234
graph = self.source.get_graph()
235
parent_map = graph.get_parent_map(revs)
236
rev_order = tsort.topo_sort(parent_map)
237
rev_id_to_root_id = self._find_root_ids(revs, parent_map, graph)
238
root_id_order = [(rev_id_to_root_id[rev_id], rev_id) for rev_id in
240
# Guaranteed stable, this groups all the file id operations together
241
# retaining topological order within the revisions of a file id.
242
# File id splits and joins would invalidate this, but they don't exist
243
# yet, and are unlikely to in non-rich-root environments anyway.
244
root_id_order.sort(key=operator.itemgetter(0))
245
# Create a record stream containing the roots to create.
246
if len(revs) > self.known_graph_threshold:
247
graph = self.source.get_known_graph_ancestry(revs)
248
new_roots_stream = _new_root_data_stream(
249
root_id_order, rev_id_to_root_id, parent_map, self.source, graph)
250
return [('texts', new_roots_stream)]
253
def _new_root_data_stream(
254
root_keys_to_create, rev_id_to_root_id_map, parent_map, repo, graph=None):
255
"""Generate a texts substream of synthesised root entries.
257
Used in fetches that do rich-root upgrades.
259
:param root_keys_to_create: iterable of (root_id, rev_id) pairs describing
260
the root entries to create.
261
:param rev_id_to_root_id_map: dict of known rev_id -> root_id mappings for
262
calculating the parents. If a parent rev_id is not found here then it
263
will be recalculated.
264
:param parent_map: a parent map for all the revisions in
266
:param graph: a graph to use instead of repo.get_graph().
268
for root_key in root_keys_to_create:
269
root_id, rev_id = root_key
270
parent_keys = _parent_keys_for_root_version(
271
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph)
272
yield versionedfile.FulltextContentFactory(
273
root_key, parent_keys, None, '')
276
def _parent_keys_for_root_version(
277
root_id, rev_id, rev_id_to_root_id_map, parent_map, repo, graph=None):
278
"""Get the parent keys for a given root id.
280
A helper function for _new_root_data_stream.
282
# Include direct parents of the revision, but only if they used the same
283
# root_id and are heads.
284
rev_parents = parent_map[rev_id]
286
for parent_id in rev_parents:
287
if parent_id == NULL_REVISION:
289
if parent_id not in rev_id_to_root_id_map:
290
# We probably didn't read this revision, go spend the extra effort
293
tree = repo.revision_tree(parent_id)
294
except errors.NoSuchRevision:
295
# Ghost, fill out rev_id_to_root_id in case we encounter this
297
# But set parent_root_id to None since we don't really know
298
parent_root_id = None
300
parent_root_id = tree.get_root_id()
301
rev_id_to_root_id_map[parent_id] = None
303
# rev_id_to_root_id_map[parent_id] = parent_root_id
304
# memory consumption maybe?
306
parent_root_id = rev_id_to_root_id_map[parent_id]
307
if root_id == parent_root_id:
308
# With stacking we _might_ want to refer to a non-local revision,
309
# but this code path only applies when we have the full content
310
# available, so ghosts really are ghosts, not just the edge of
312
parent_ids.append(parent_id)
314
# root_id may be in the parent anyway.
316
tree = repo.revision_tree(parent_id)
317
except errors.NoSuchRevision:
318
# ghost, can't refer to it.
322
parent_ids.append(tree.get_file_revision(root_id))
323
except errors.NoSuchId:
326
# Drop non-head parents
328
graph = repo.get_graph()
329
heads = graph.heads(parent_ids)
331
for parent_id in parent_ids:
332
if parent_id in heads and parent_id not in selected_ids:
333
selected_ids.append(parent_id)
334
parent_keys = [(root_id, parent_id) for parent_id in selected_ids]
338
class TargetRepoKinds(object):
339
"""An enum-like set of constants.
341
They are the possible values of FetchSpecFactory.target_repo_kinds.
344
PREEXISTING = 'preexisting'
349
class FetchSpecFactory(object):
350
"""A helper for building the best fetch spec for a sprout call.
352
Factors that go into determining the sort of fetch to perform:
353
* did the caller specify any revision IDs?
354
* did the caller specify a source branch (need to fetch its
355
heads_to_fetch(), usually the tip + tags)
356
* is there an existing target repo (don't need to refetch revs it
358
* target is stacked? (similar to pre-existing target repo: even if
359
the target itself is new don't want to refetch existing revs)
361
:ivar source_branch: the source branch if one specified, else None.
362
:ivar source_branch_stop_revision_id: fetch up to this revision of
363
source_branch, rather than its tip.
364
:ivar source_repo: the source repository if one found, else None.
365
:ivar target_repo: the target repository acquired by sprout.
366
:ivar target_repo_kind: one of the TargetRepoKinds constants.
370
self._explicit_rev_ids = set()
371
self.source_branch = None
372
self.source_branch_stop_revision_id = None
373
self.source_repo = None
374
self.target_repo = None
375
self.target_repo_kind = None
378
def add_revision_ids(self, revision_ids):
379
"""Add revision_ids to the set of revision_ids to be fetched."""
380
self._explicit_rev_ids.update(revision_ids)
382
def make_fetch_spec(self):
383
"""Build a SearchResult or PendingAncestryResult or etc."""
384
if self.target_repo_kind is None or self.source_repo is None:
385
raise AssertionError(
386
'Incomplete FetchSpecFactory: %r' % (self.__dict__,))
387
if len(self._explicit_rev_ids) == 0 and self.source_branch is None:
388
if self.limit is not None:
389
raise NotImplementedError(
390
"limit is only supported with a source branch set")
391
# Caller hasn't specified any revisions or source branch
392
if self.target_repo_kind == TargetRepoKinds.EMPTY:
393
return vf_search.EverythingResult(self.source_repo)
395
# We want everything not already in the target (or target's
397
return vf_search.EverythingNotInOther(
398
self.target_repo, self.source_repo).execute()
399
heads_to_fetch = set(self._explicit_rev_ids)
400
if self.source_branch is not None:
401
must_fetch, if_present_fetch = self.source_branch.heads_to_fetch()
402
if self.source_branch_stop_revision_id is not None:
403
# Replace the tip rev from must_fetch with the stop revision
404
# XXX: this might be wrong if the tip rev is also in the
405
# must_fetch set for other reasons (e.g. it's the tip of
406
# multiple loom threads?), but then it's pretty unclear what it
407
# should mean to specify a stop_revision in that case anyway.
408
must_fetch.discard(self.source_branch.last_revision())
409
must_fetch.add(self.source_branch_stop_revision_id)
410
heads_to_fetch.update(must_fetch)
412
if_present_fetch = set()
413
if self.target_repo_kind == TargetRepoKinds.EMPTY:
414
# PendingAncestryResult does not raise errors if a requested head
415
# is absent. Ideally it would support the
416
# required_ids/if_present_ids distinction, but in practice
417
# heads_to_fetch will almost certainly be present so this doesn't
419
all_heads = heads_to_fetch.union(if_present_fetch)
420
ret = vf_search.PendingAncestryResult(all_heads, self.source_repo)
421
if self.limit is not None:
422
graph = self.source_repo.get_graph()
423
topo_order = list(graph.iter_topo_order(ret.get_keys()))
424
result_set = topo_order[:self.limit]
425
ret = self.source_repo.revision_ids_to_search_result(result_set)
428
return vf_search.NotInOtherForRevs(self.target_repo, self.source_repo,
429
required_ids=heads_to_fetch, if_present_ids=if_present_fetch,
430
limit=self.limit).execute()
101
self.count_weaves = 0
102
self.copied_file_ids = set()
104
self.pb = bzrlib.ui.ui_factory.progress_bar()
107
self.from_branch.lock_read()
109
self._fetch_revisions(last_revision)
111
self.from_branch.unlock()
114
def _fetch_revisions(self, last_revision):
115
self.last_revision = self._find_last_revision(last_revision)
116
mutter('fetch up to rev {%s}', self.last_revision)
117
if (self.last_revision is not None and
118
self.to_branch.has_revision(self.last_revision)):
121
revs_to_fetch = self._compare_ancestries()
123
raise InstallFailed([self.last_revision])
124
self._copy_revisions(revs_to_fetch)
125
self.new_ancestry = revs_to_fetch
127
def _find_last_revision(self, last_revision):
128
"""Find the limiting source revision.
130
Every ancestor of that revision will be merged across.
132
Returns the revision_id, or returns None if there's no history
133
in the source branch."""
136
self.pb.update('get source history')
137
from_history = self.from_branch.revision_history()
138
self.pb.update('get destination history')
140
return from_history[-1]
142
return None # no history in the source branch
145
def _compare_ancestries(self):
146
"""Get a list of revisions that must be copied.
148
That is, every revision that's in the ancestry of the source
149
branch and not in the destination branch."""
150
self.pb.update('get source ancestry')
151
self.from_ancestry = self.from_branch.get_ancestry(self.last_revision)
153
dest_last_rev = self.to_branch.last_revision()
154
self.pb.update('get destination ancestry')
156
dest_ancestry = self.to_branch.get_ancestry(dest_last_rev)
159
ss = set(dest_ancestry)
161
for rev_id in self.from_ancestry:
163
to_fetch.append(rev_id)
164
mutter('need to get revision {%s}', rev_id)
165
mutter('need to get %d revisions in total', len(to_fetch))
166
self.count_total = len(to_fetch)
169
def _copy_revisions(self, revs_to_fetch):
171
for rev_id in revs_to_fetch:
175
if self.to_branch.has_revision(rev_id):
177
self.pb.update('fetch revision', i, self.count_total)
178
self._copy_one_revision(rev_id)
179
self.count_copied += 1
182
def _copy_one_revision(self, rev_id):
183
"""Copy revision and everything referenced by it."""
184
mutter('copying revision {%s}', rev_id)
185
rev_xml = self.from_branch.get_revision_xml(rev_id)
186
inv_xml = self.from_branch.get_inventory_xml(rev_id)
187
rev = serializer_v5.read_revision_from_string(rev_xml)
188
inv = serializer_v5.read_inventory_from_string(inv_xml)
189
assert rev.revision_id == rev_id
190
assert rev.inventory_sha1 == sha_string(inv_xml)
191
mutter(' commiter %s, %d parents',
194
self._copy_new_texts(rev_id, inv)
195
parents = rev.parent_ids
196
new_parents = copy(parents)
197
for parent in parents:
198
if not self.to_branch.has_revision(parent):
199
new_parents.pop(new_parents.index(parent))
200
self._copy_inventory(rev_id, inv_xml, new_parents)
201
self.to_branch.revision_store.add(StringIO(rev_xml), rev_id)
202
mutter('copied revision %s', rev_id)
204
def _copy_inventory(self, rev_id, inv_xml, parent_ids):
205
self.to_control.add_text('inventory', rev_id,
206
split_lines(inv_xml), parent_ids,
207
self.to_branch.get_transaction())
209
def _copy_new_texts(self, rev_id, inv):
210
"""Copy any new texts occuring in this revision."""
211
# TODO: Rather than writing out weaves every time, hold them
212
# in memory until everything's done? But this way is nicer
213
# if it's interrupted.
214
for path, ie in inv.iter_entries():
215
if ie.revision != rev_id:
217
mutter('%s {%s} is changed in this revision',
219
self._copy_one_weave(rev_id, ie.file_id)
221
def _copy_one_weave(self, rev_id, file_id):
222
"""Copy one file weave."""
223
mutter('copy file {%s} modified in {%s}', file_id, rev_id)
224
if file_id in self.copied_file_ids:
225
mutter('file {%s} already copied', file_id)
227
from_weave = self.from_weaves.get_weave(file_id,
228
self.from_branch.get_transaction())
229
to_weave = self.to_weaves.get_weave_or_empty(file_id,
230
self.to_branch.get_transaction())
232
to_weave.join(from_weave)
233
except errors.WeaveParentMismatch:
234
to_weave.reweave(from_weave)
235
self.to_weaves.put_weave(file_id, to_weave,
236
self.to_branch.get_transaction())
237
self.count_weaves += 1
238
self.copied_file_ids.add(file_id)
239
mutter('copied file {%s}', file_id)
added
removed
bzrlib/fetch.py
